shoeprint

Community Help
1

I am working on a project related to shoes. My goal is to develop a model that can classify geometric patterns (such as circles, triangles, quadrilaterals, letters, and logos) found on shoe soles. I have a dataset of 10,000 shoe sole images for this project. I would like to use Roboflow for this purpose. Can you recommend a suitable approach for me? Thank you in advance for your support.

2

Hi @ramazan_oguz ,

Thank you for considering Roboflow for your project!

So would your workflow look like this:

  1. scan photo with shoe sole
  2. get list of shapes on the sole (in a form: [‘star’, ‘rectangle’, ‘triangle’])
  3. pass that list to other component of your app that would make use of this information

Or:

  1. scan photo with shoe sole
  2. get all instances of detected shapes on the sole (i.e. [{"class": "triangle", "xyxy": [10, 10 20, 20]}, {"class": "triangle", "xyxy": [30, 30, 40, 40]}, {"class": "star", "xyxy": [50, 50, 60, 60]}])
  3. pass that list to other component of your app that would make use of this information

Or:

  1. scan photo with shoe sole
  2. get all pixel-perfect outlines of detected shapes (i.e. [{"class": "triangle", "polygon": [...]}, {"class": "triangle", "polygon": [...]}, {"class": "star", "polygon": [...]}])
  3. pass that list to other component of your app that would make use of this information

First option (multi-modal classification) would be probably fastest, last option (segmentation) would be slowest but with most amount of information (also, given pixel-perfect outlines you can calculate bounding boxes). 2nd option (object-detection) is middle-ground between 1st and 3rd option.

Hope this helps,
Grzegorz